INNER CODE UNIT · Python

next_prob

FoundationVision/Liquid · evaluation/app.py:252

                next_token, next_prob = sample(next_token_logits, **sampling_kwargs)
                pred_tokens.append(next_token)

            # update generated ids, model inputs, and length for next step
            input_ids = torch.cat([input_ids, next_token], dim=-1)
            model_kwargs = vqllm._update_model_kwargs_for_generation(
                outputs,
                model_kwargs,
                is_encoder_decoder=vqllm.config.is_encoder_decoder,
            )

        del sampling_kwargs
        del model_inputs
        del outputs
        image_vq_id = torch.cat(pred_tokens,dim=1)-ori_vocabe_size
        image_vq_id = torch.clamp(image_vq_id, min=0, max=8191)
        
        generated_image_list = []

View source record →